About
Timbaly is built in Turin, Italy, on one observation: most small-business websites are not badly built, they are badly maintained. The site ships, everyone is pleased, and then nothing happens to it for two years.
That is not a discipline problem. It is what happens when the work that decides whether you get found depends on somebody remembering to do it every month.
The name
From the timbal — a struck-string instrument. The mark is a hammer above a vibrating string.
The strings are the channels a business can be found on: Google, Bing, ChatGPT, Claude, Perplexity, Gemini. Each one is tuned differently and each one goes out of tune on its own. Keeping them in tune is not a launch task, it is a standing one — and that is the whole product in a metaphor.
It also explains the design. No stock photography, no gradients, no rounded cards: 1px rules, a brass accent, and typography. A site about being legible to machines that is itself illegible would be a poor advertisement.
Principles
Any company can list values. These are the four that changed the product, listed with the price we pay for each — because a principle with no cost attached is just marketing.
No generative engine publishes a citation metric, so we do not sell one. Where a figure does not exist, the tools say so instead of returning zero — because a model reading “zero clicks” concludes the site gets no traffic, which is a false statement said with confidence.
What it costs: the single most saleable dashboard in this market is one we will not build. Competitors show a rising AI-visibility line. We show the reasoning instead.
The Google Search Console property is created inside the customer's own Google account, in their name, and stays theirs with sixteen months of history if they leave. We hold read permission, revocable at any time.
What it costs: an OAuth flow, a signed state, encrypted refresh tokens, a verification state machine and a category of support question we would not otherwise have. The convenient version would have been one platform-owned property per site.
Privacy notices come from fixed text, not generation. The consent checkbox is inserted by code. Mobile layout faults are corrected deterministically. Structured data, canonical tags and the root files are applied by the publishing pipeline.
What it costs: considerably more code than “ask the model nicely”. The prompt already produces the right thing almost every time — but on a platform that publishes by itself, almost means some customers go live unlawfully.
Google cannot be asked to index a page. Small sites have no field speed data. Generation fails on roughly one page in ten. The free plan is noindex. The admin panel is not white-labelled. All of it is on the public pages.
What it costs: conversions, certainly. It buys the thing we would rather have: a customer who is not surprised in month two, and pages a careful reader — or a careful model — is willing to quote.
Practising it
Every page here was written by an AI assistant over the same MCP connector customers use, and marked as AI-generated for EU AI Act art. 50. Not a mock-up of the product — the product.
English at the root, Italian under /it, linked with hreflang, each with its own localized slugs rather than duplicated URLs.
robots.txt naming every AI crawler, llms.txt, sitemap.xml, structured data on every page. You can check all of them right now.
Where we are
We are a small team, which is relevant in two directions. It is why support is a person and not a ticket queue, and it is also why some things on the roadmap are honestly not there yet — a white-labelled admin panel, per-site team roles, sampling for AI visibility.
If you want to know whether Timbaly fits a specific situation, the fastest route is to ask. The contact page takes one message, and the answer will include a “no, use something else” if that is the true answer.
Three minutes, no card, and you are looking at real pages about your own business rather than a page about our principles.